What Should We Do With These? Challenges related to (semi-)automatically detected sites and features. A note

Abstract

Recent advances in machine learning and computer vision techniques have brought (semi-)automatic feature detection within reach of an increasing number of archaeologists and archaeological institutions, including those in Finland. These techniques improve our ability to detect and gather information on archaeological cultural heritage over vast areas in a highly efficient manner. However, the widespread adoption of such methods can also pose significant challenges for archaeological cultural heritage management, especially in relation to certain types of near-ubiquitous archaeological remains from the 17th-20th centuries. In general, machine learning based methods are especially well suited for detecting features that have relatively uniform characteristics, are present in sufficiently high numbers, and are easily discernible in remote-sensing datasets. In Finland, most archaeological features that meet these criteria are relatively recent features, such as tar and charcoal kilns from the 17th-20th centuries or remains of World War I and II era defensive structures. Although there are some exceptions to this rule, such as prehistoric pitfall trap systems, the archaeological features selected by the use of (semi-)automatic detections is heavily skewed towards only a handful of relatively recent feature types. While this is not necessarily a problem per se, it poses a series of questions for the cultural heritage management sector which have sparked active discussion. This article seeks to summarise and reflect on some of these perspectives.

Keywords

• machine learning
• deep learning
• lidar
• cultural heritage management
• archaeology
• finland
• airborne laser scanning